How to Use Custom Model Architectures?

[DylanZelkin]

I am using the skrl train script with my custom environment, but now I would like to use a custom model architecture instead of just using the yml config file to chose between the premade options. I cannot find any tutorials to do so, is this not possible right now? I would prefer not to meddle with remaking the entire training workflow and scripts, will it require significant source code changes, or can i do this relatively easily, and what are the steps to do so?

please help, thanks.

[Toni-SM]

Hi @DylanZelkin

You can refer to the skrl's example docs for standalone python script that allow you to implement custom models beyond the limitation the yaml file has.

https://skrl.readthedocs.io/en/latest/intro/examples.html#nvidia-isaac-lab